my problem is that I want to filter a DataFrame to only include times within the interval [start, end) . If do not care about the day, I would like to filter only for start and end time for each day. I have a solution for this but it is slow. So my question is if there is a faster way to do the time based filtering.
Example
import pandas as pd
import time
index=pd.date_range(start='2012-11-05 01:00:00', end='2012-11-05 23:00:00', freq='1S').tz_localize('UTC')
df=pd.DataFrame(range(len(index)), index=index, columns=['Number'])
# select from 1 to 2 am, include day
now=time.time()
df2=df.ix['2012-11-05 01:00:00':'2012-11-05 02:00:00']
print 'Took %s seconds' %(time.time()-now) #0.0368609428406
# select from 1 to 2 am, for every day
now=time.time()
selector=(df.index.hour>=1) & (df.index.hour<2)
df3=df[selector]
print 'Took %s seconds' %(time.time()-now) #Took 0.0699911117554
As you can see if I remove the day (second case) it takes almost twice as much. The computation time increases rapidly if I have a number of different days, e.g from 5 to 7 Nov:
index=pd.date_range(start='2012-11-05 01:00:00', end='2012-11-07 23:00:00', freq='1S').tz_localize('UTC')
So, to summarize is there a faster method to filter by time of the day, across many days?
Thx
解决方案
You need between_time method.
In [14]: %timeit df.between_time(start_time='01:00', end_time='02:00')
100 loops, best of 3: 10.2 ms per loop
In [15]: %timeit selector=(df.index.hour>=1) & (df.index.hour<2); df[selector]
100 loops, best of 3: 18.2 ms per loop
I had done these tests with 5th to 7th November as index.
Documentation
Definition: df.between_time(self, start_time, end_time, include_start=True, include_end=True)
Docstring:
Select values between particular times of the day (e.g., 9:00-9:30 AM)
Parameters
----------
start_time : datetime.time or string
end_time : datetime.time or string
include_start : boolean, default True
include_end : boolean, default True
Returns
-------
values_between_time : type of caller